1

Load Bearing Capacity of Tempered Structural

Glass
By Hlne Carr
1
and L. Daudeville
2
ABSTRACT: This paper presents a method for the failure analysis of structural glass components of
buildings. Structural glass is generally prestressed by tempering. In the method, residual stresses are
first computed by simulation of this tempering process of soda-lime-silica glass plates using the finite
element method. The model includes both stress and structural relaxations. The edge effects of
tempering are modeled. Then, the failure strength of annealed glass is obtained by a statistical
analysis of tests performed on small specimens. Loading rate effects are also taken into account.
Computational results are associated with the statistical approach for the failure prediction of large
annealed and tempered glass plates. This prediction method is validated by experimental results from
four-point bending tests up to failure.
KEY WORDS: structural glass, thermal tempering, finite elements, fracture, loading rate effect, size
effect, Weibull model, failure tests
2
INTRODUCTION
New applications of tempered glass in structural parts of buildings (posts, beams, shear walls)
necessitate a good knowledge of the load carrying capacity and the life-time of structural glass
components. The objective of this work is the strength prediction of tempered soda-lime-silica glass
plates loaded in plane.
Tempered glass may be regarded as a prestressed material since its thermal treatment induces a
certain amount of residual stresses. As these stresses are hardly measurable at all points of an
element, simulation of the tempering process is necessary in order to evaluate the transient and
residual stress states. Previous studies of tempering have been concerned with the calculation of
residual stresses in infinitely thin plates (Narayanaswamy and Gardon 1969; Gardon 1980; Burke et
al. 1987; Carr and Daudeville 1996). Knowledge of the three-dimensional stress state in the whole
plate is necessary since catastrophic failure is the consequence of the propagation of cracks
originated by machining the edges of glass plates (the influence of volume defects being neglected).
Because of the present scarcity of knowledge of the state of residual stresses near straight edges or
holes, full-scale tests are generally necessary to design tempered glass elements. The three-
dimensional calculation of residual stresses near edges has not been shown before.
In this paper, computational results of the three-dimensional Finite Element (FE) simulation of
tempering of a thick glass plate are first presented. Since failure properties of glass during the various
stages of solidification are unknown, transient stresses have possible application only in controlling
the quenching process and the resulting residual stresses.
Then, experimental results of Four-Point Bending (FPB) tests performed on small annealed glass
specimens are shown. The strength distribution and static fatigue sensitivity are determined from tests
3
under constant stress rates. The influences of the loading rate and surface finish are investigated.
Tests were performed in usual moisture conditions (i.e. the influence of the relative humidity was not
studied).
For a reliable failure analysis of glass elements, a Weibull model is identified from tests performed on
the small annealed specimens. The loading rate effect is included in the Weibull model by means of
fracture mechanics concepts.
The goal of the paper is strength prediction of large structural elements. The origin of fracture in
structural glass is via mechanical damage due to machining that involves abrading the surface. The
initial size of edge scratches is between 10 and 100 m. Their final size before catastrophic
propagation is of the same order of magnitude for annealed glass or tempered glass under flexion.
The stress state of tempered plates under bending can then be considered as quasi-uniform in the
vicinity of edge cracks. The analysis is different from the problem of non-loaded tempered glass
objects (notably automobile windscreens) containing clearly visible surface cracks in which the
variation of stresses along the crack must be taken into account.
A superposition model is proposed for the strength prediction of tempered glass plates. The model
uses results from the numerical simulations of tempering, and data from the probabilistic model of the
studied glass with machined cracks. The main assumptions can be debated, but the simplified model
must be considered as a first step. Some FPB tests up to failure were performed on large annealed
and tempered elements. Experiments on annealed specimens confirm the ability of the statistical
model to describe the size effect. Another original aspect of the paper is the use of optical
measurements on large tempered specimens which allow the identification of unknown heat transfer
parameters of the tempering process and the validation of the FE model. Failure tests validate the
superposition method and the Weibull model.
4
THERMAL TEMPERING SIMULATION
Thermal tempering of glass consists of cooling very quickly, by air casts, a plate which has been
heated to approximately 620C. This treatment confers a strengthening by means of a residual stress
state of tension in the core and of compression near the surface. The goal of FE simulations of
quenching is to obtain residual stresses of thick tempered soda-lime-silica glass plates.
The present difficulty of such a problem is the correct modeling of heat transfers especially in the
case of a complex geometry (e.g. a holed plate). Obtaining reliable data of the thorough
Narayanaswamy's model (1978) used in this study is another difficulty. Available data in this paper
are issued from experimental studies carried out at Saint-Gobain Recherche.
Thermomechanical behavior of glass
The behavior varies quickly around the transition temperature (T
g
580C) between the glass and
liquid states. The presented model includes stress relaxation due to viscosity and structural
relaxation due to the actual state of structure of glass. The viscous behavior can be neglected at
20C.
Mechanical behavior
Temperature is first considered constant. The mechanical behavior of glass is described in terms of
stress relaxation by means of a generalized Maxwell model (Gy et al. 1994; Duffrene 1994).
Relaxation shear and bulk moduli are described with instantaneous and deferred moduli and
expanded into Pronys series. The deferred shear modulus value is zero:

G t () 2G
g

1
t () (1)

K t ( ) 3K
e
3K
e
3K
g
( )

2
t () (2)
5

i
t
()
w
ij
exp
t

ij



_
,

j1
n
i

i = 1, 2 (3)
Influence of temperature
A relaxation function

i
T
( ) ( )
known at the reference temperature (T
ref
) can be determined at any
temperature (T) with a classical time-temperature equivalence (Schwarzl and Staverman 1952) by
means of the reduced time () defined as:

i
T,t ( )
i
T
ref
, ( ) i = 1, 2 (4)
Weights and relaxation times, defined in (3), are assumed constant with temperature. An Arrhenius
relation allows a correct description of the dependency of relaxation times with temperature
(Narayanaswamy 1978; Gardon 1980).

t, T ( ) exp
H
R
1
T
ref

1
T(t' )



_
,




_
,

0
t

dt' (5)
Structural relaxation
During the quenching, and for temperatures close to the transition range, the glass structure can not
be stabilized. There are several possible glassy states for one temperature depending on the cooling
rate (Fig. 1). This is the structural relaxation (Narayanaswamy and Gardon 1969; Narayanaswamy
1971).
The fictitious temperature (T
f
) was introduced to account for the structure of glass (Tool 1946). A
structural relaxation volume function (M
v
) is defined as (Fig. 1):

M
v
t ( )
V t
( )
V
, 2
V
0, 2
V
, 2

T
f
T
2
T
1
T
2
(6)
6
Its temperature dependency is taken into account with the reduced time ((t)). The fictitious
temperature (T
f
) is defined as:

T
f
t () T t ( ) M
V
t () t ( )
[ ]
dT t
( )
d t
d t
0
t

(7)
By analogy with the viscous relaxation, the response function can be described with a Prony's series:

M
v
( ) C
i
exp

i



_
,

i 1
n

(8)
Structural relaxation times (
i
) are considered to be proportional to shear relaxation times (Guillemet
et al. 1992).
The dependency of viscosity with the structure state is not considered in the presented FE
simulations since it induces very small variations of residual stresses (Gardon 1980). But the
dependency of the structural state with the density, which is much more significant on residual
stresses, is introduced by means of the thermal expansion coefficient variations with the temperature.

th

g
T t
()
T
f
t
() ( )
+
l
T
f
t
( )
T
0 ( )
(9)
Glass characteristics
The thermoviscoelastic characteristics (Duffrenne 1994) are given in Table 1 and Table 2. The
structural relaxation characteristics (Guillemet et al. 1992) are in Table 2.
The thermal conductivity and the specific heat (Guillemet et al. 1992) are characteristics varying with
temperature, in Table 3.
7
Finite element analysis
Previous works investigated inner residual stresses of thin plates (thickness 0.61 cm)
(Naranayaswamy 1978; Gardon 1980; Burke et al. 1987; Carr and Daudeville 1996). This work
concerns thick tempered glass plates of building structures loaded in plane. The origin of fracture is
located on the plate edges. Therefore, transient and residual stresses are analyzed in both inner and
edge zones with the FE code MARC. All of the following results are related to large tempered plates
(2000x300x19 mm
3
).
Mesh and boundary conditions
Although the problem does not depend on z, three-dimensional elements are used to account for
thermal strain along this coordinate (Fig. 2). The mesh is refined in zones of high stress gradients, in
the plate thickness (along x) and close to the edge (along y).
The thermal treatment is assumed to be uniform on all plate faces. The mechanical boundary
conditions are symmetry conditions (i.e. only one eighth of the plate is modelled). The thermal
boundary conditions are forced convections to model the blowing by air casts (air temperature T
ext
=
20C). Three constant heat transfer coefficients (h
1
, h
2
= h
3
) are defined. The plate temperature is
assumed to be uniform and equal to T
0
(620C) at the beginning of cooling.
The problem to solve is a time-dependent problem because of the viscous behavior of glass.
Identification of heat transfer coefficients
The only unknown parameters of the tempering process simulation are the two heat transfer
coefficients h
1
and h
2
. They are identified by means of comparisons between simulation results and
optical measurements (Aben and Guillemet 1993; Sinha 1978; Redner 1994) of residual stresses
performed on large tempered plates.
8
The epibiascope is used to obtain the surface stress (
yy
=
zz
) in several points in the inner part of
the plate. This measurement allows the identification of h
1
(Table 4).
The Babinet compensator measures the difference of the optical path in the thickness of the plate
which is proportional to the integral in the thickness (t) of the difference of principal stresses
(

1
t
(
yy
t / 2
+ t / 2

zz
)dx). One measurement is carried out close to the edge to identify h
2
(Table 4,
at 1.5 mm because of the chamfer, point A of Fig. 3).
Several measures at different distances from the edge were carried out. The comparison between
measurements and numerical simulations with the distance from the edge are given in Fig. 3. The
good comparison validates the FE calculations of residual stresses.
Transient and residual stresses
At the beginning of cooling, the surface contracts more quickly than the core. By equilibrium, the
core is under compression and the surface under tension. The surface tension, particularly important
at the beginning of cooling at the plate corner, may lead to the fracture of the tempered glass plate.
As the surface temperature becomes less than the transition temperature, the surface freezes and the
still liquid core continues to contract. Then, the surface becomes compressed while the core is under
tension, by equilibrium (Fig. 4).
In the inner part of the plate, the residual stress (
yy
=
zz
) has a parabolic shape in the plate
thickness. The residual edge stress (
zz
) is quasi-constant in the thickness and slightly less than the
inner surface stress (Fig. 5).
9
FRACTURE PARAMETERS IDENTIFICATION
The failure stress of a macroscopic annealed glass element under tension is between 30 and 100
MPa, whereas the tensile strength of a glass fiber is about 20 GPa. Griffith (1920) explained this by
the presence of micro-cracks. The origin of fracture is located on the machined edges. Failure is
governed on one hand by their propagation and on the other hand by their random distribution.
Statistical model
The Weibull model (1951) is a statistical approach for the failure analysis of brittle materials with
random defects. The failure probability (P
f
) of a glass plate is:

P
f
1 exp
1
S
0

u

0



_
,

m
S

1
]
1
1
(10)
This model accounts for the size effect and the stress concentration effect. Weibull parameters
depend on the material but also on the loading rate because of the subcritical crack growth.
For a glass plate under bending, S is the polished surface under tension. The integration of (10) on S
is then explicit. It was verified for each test presented in the next sections that cracking had originated
on S and not below the surface (for annealed or tempered glass specimens).
Subcritical crack growth
Glass strength depends on the rate and the duration of loading. This phenomenon, also named static
fatigue, was shown by Grenet (1899). This is not observed in vacuum conditions and it is due to the
effect of moisture at the crack tips (Michalske and Freiman 1983).
Failure due to the propagation of cracks from the edge can be modeled by means of the stress
intensity factor (K
I
) of fracture mechanics. Glass is a brittle material and so the fracture mode can be
10
considered as a pure mode I. The crack velocity depends on K
I
(Wiederhorn 1967) because of the
subcritical crack growth. Evans (1974) proposed the following model:
K
I
< K
Ith
no crack propagation
K
Ith
= K
I
< K
Ic
subcritical crack growth:

da
dt
= A K
I
n
(11)
K
Ic
= K
I
< K
Icb
crack propagation
K
I
= K
Icb
crack branching
The stress intensity factor is calculated with the applied stress (far away from the crack) :

K
I
Y a (12)
The shape factor value can be issued from (Rooke and Cartwright 1976).
Identification tests
Description of the tests
Small annealed specimens with polished edges were tested under four point bending with different
loading rates (Fig. 6). The influences of both the loading rate and the surface finish were studied. A
first set of specimens was tested at 0.05 and 0.5 MPa/s under displacement control. Then a second
set of specimens, with a different surface finish, was tested at 0.05, 0.5 and 5 MPa/s.
The reference surface finish is the first one since large specimens described further were machined
with this surface finish.
On Fig. 7, circles and triangles give, for the reference surface finish and for all the tested specimens,
the experimental cumulative failure probability at 0.05 and 0.5 MPa/s with respect to the applied
stress on the tensile edge. It can be observed, first, that there is a large scattering of the failure stress
justifying a statistical approach, and second, that there is an influence of the loading rate.
11
Table 5 gives the experimental mean failure stress and the two experimental stresses corresponding
to failure probabilities of 0.2 and 0.8. Results are very different for the two qualities of machining.
Loading rate effect
Association of the Weibull and Evans models enables one to account for the subcritical crack growth
in the statistical analysis. The apparent Weibull parameters (

0
, m) depend on both the loading rate
and the environment for a given surface finish. Intrinsic parameters (m*,
0
*) are defined from the
strength in inert medium.
Assuming
u
= 0 and S
0
= 1, (10) gives for a FPB test in inert medium:

P
f
1 exp S

0
*



_
,

m
*




1
]
1
1
(13)
Intrinsic Weibull parameters (

0
*
, m
*
) are independent of the loading conditions. The failure
probability for a constant stress or stress rate can be written with the intrinsic parameters and the
subcritical crack growth parameters (Helfinstine 1980; Carr 1996).
Assuming the initial defect length is small compared to the final crack length (a(t = 0) << a(t = t
f
)) and
that the value of the threshold stress intensity factor is zero (K
Ith
= 0), the integration of (11) in time,
(12) and (13) give:

P
f
1 exp S

0



_
,

m




1
]
1
1
1 exp S

_
,

nm

n2
t
f
t
*



_
,

m

n 2
k(n)





1
]
1
1
1
(14)
With :
12

t
*

2
n 2
1
AY
2

2
K
Ic
2 n
(15)
The factor k(n) depends on the loading rate:
k(n) = 1 if


0 and k(n) =

1
n + 1



_
,

m

n2
if


constant (16)
Then, the apparent Weibull parameters are:

m
(n +1) m
*
n 2
(17)

0

0
*

_
,
n 2
n + 1


2(n +1)
(n 2)AY
2
K
Ic
n2



_
,

1
n
(18)
Only the apparent Weibull stress (
0
) depends on the stress rate. If the initial crack length is not
neglected, the two apparent Weibull parameters (m,
0
) depend on the loading rate (Carr 1996).
Determination of Weibull parameters
Apparent Weibull parameters (m,
0
)
0.5
were identified with the test results shown in Fig. 7 obtained
at a stress rate of 0.5 MPa/s. The maximum likelihood method was used for the identification
(continuous line in Fig. 7).
The apparent Weibull parameters at 0.05 MPa/s can be calculated with (17) and (18). It should be
noticed that the knowledge of fracture mechanics parameters A, Y and K
Ic
is not necessary since the
calculation of the intrinsic Weibull stress (

0
*
) is not carried out:

m ( )
0.05
m ( )
0.5
and
0
( )
0.05

0
( )
0.5
0.05
0.5



_
,

1
n
(19)
13
The failure prediction at 0.05 MPa/s is given by a dashed line in Fig. 7. The correct prediction
validates the association of the subcritical crack growth model with the Weibull theory. The only
fracture mechanics parameter used in this study is the exponent (n). This material parameter is
difficult to obtain since it requires long term observations of crack propagation. For usual moisture
conditions in buildings, n was identified from double torsion tests (n = 12.76, private communication
from Saint-Gobain Recherche).
STRENGTH PREDICTION OF TEMPERED GLASS ELEMENTS
In one case, numerical simulations of glass tempering have given the residual stress state of thick
plates. In another case, tests on small annealed glass samples have allowed the identification of
statistical failure parameters of the studied glass.
The strength prediction of tempered glass elements will now be proposed by associating the two
previous analysis.
The superposition method
A key point is that the crack length before failure can be considered small compared to the lengths of
stress variation. Fig. 8 gives the residual stress parallel to the edge with respect to the edge distance
in the mid-plane of a large tempered plate. The zoomed image shows that the residual stress can be
considered as constant along a distance which is about one crack length before failure (about 100
m) (Carr 1996).
The study of a clearly visible crack would involve taking into account the variation of stress along the
crack length (Lawn and Marshall 1977). Here, in the studied problem, the crack does not grow
through a region of negative K
I
(due to a residual stress of compression) to a region of positive K
I
14
for a long crack length since the crack will not propagate as long as the whole crack is in a region
under compression.
The stress state, that is the superposition of residual and bending stresses, can be considered as
uniform far away from the crack. Then, a crack propagation will be possible when the sum of the
bending and residual stress values reaches a critical value.
The strength prediction is based on two major assumptions:
Residual stresses due to tempering are deterministic data. This assumption relies on the
observation of a small dispersion shown from optical measurements of residual stresses on
several points of different glass plates tempered in the same conditions (Table 4).
The surface flaw distribution is not affected by tempering in spite of a surface tension stress at
the beginning of the cooling process which may induce a crack propagation.
The superposition method gives (at the location of failure):
tempered glass strength (T) = annealed glass strength (A) + residual stress (R) (20)
(Weibull model) (FE simulation)
Tests on large glass elements
Large annealed and tempered glass plates (2000x300x19mm
3
) were tested under FPB at 0.05
MPa/s (Fig. 9). A system prevents the plates from lateral bending. It was verified that this system
does not induce any out-of-plane bending stresses with gauges on each face of the plate. The edges
of large and small elements were machined in the same conditions (first surface finish).
15
Annealed specimens
In a first step, the Weibull model is used for the failure probability prediction of large annealed glass
plates. The aim is the verification of the size effect description. The strength predictions for three
probabilities of failure are obtained with eqs. (14)-(19) (term A in (20)) and are given in Table 6.
The size-effect is clearly shown by comparing Tables 5 and 6.
According to the model, the mean experimental failure stress (
exp
) corresponds to a failure
probability of 0.26. The small difference (5%) between the mean experimental and predicted failure
stresses may be due to the influence of n (eq. (19)) whose value is difficult to obtain.
Tempered specimens
In a second step, the strength prediction of tempered glass plates (term T in (20)) is compared to
experimental results. The minimum residual compression stress obtained on the edge by FE
calculation (term R in (20)) is 96 MPa (Fig. 5). This stress was obtained near the chamfer.
Calculated failure stresses are given in Table 6.
The difference between the mean experimental and calculated failure stress of tempered plates (2 %)
may be due to the influence of n but also on the uncertainty in the identification of the heat transfer
coefficients (Table 4) due to the apparatus accuracy.
CONCLUSIONS
The FE simulation of thermal tempering of a soda-lime-silica glass plate was presented. Previous
studies have not considered edge effects of tempering. A three-dimensional analysis is carried out in
the vicinity of the edge plate where optical measurements can not give direct accurate estimations of
residual stresses. It was proposed to identify the unknown heat transfer coefficients by means of a
few optical measurements on the edge and in the center part of the plate that give some information
16
on residual stresses in these regions. This approach can be applied to more complex geometries such
as holes.
Tests on small annealed glass samples allow the determination of statistical annealed glass failure
parameters and pointed out the loading rate effect. The probabilistic model of Weibull is used in
association with a subcritical crack growth model. For a given surface finish, the model can take into
account the effects of the specimen size, of the stress distribution and of the rate of loading.
With the results of both previous models, the superposition method allows the strength prediction of
large tempered glass elements. This method is validated by the comparison between experimental
and calculated failure strengths.
The extrapolation of presented results to structural glass elements of buildings loaded on a long
period must be validated. Long term FPB tests are now in process at CSTB.
ACKNOWLEDGMENTS
The authors would like to thank J. Mazars of ENS Cachan and C. Balloche from CSTB for giving
advice and R. Gy from Saint-Gobain Recherche for advice and specimens.
APPENDIX I. REFERENCES
Aben, H. and Guillemet, C. (1993). Photoelasticity of glass. Springer Verlag, Berlin, Germany.
Burke, M.A., Soules, T.F., Busbey, R.F. and Kheson, S.M. (1987). "Finite-element calculation of
stresses in glass parts undergoing viscous relaxation." J. Am. Ceramic Society, 70 (2), 90-95.
Carr, H. (1996). "Etude dun matriau fragile prcontraint: le verre tremp." Ph.D. Thesis (in
French), Ecole Nationale des Ponts et Chausses, Champs-sur-Marne, France.
17
Carr, H. and Daudeville, L. (1996). "Numerical simulation of soda-lime silicate glass tempering." J.
de Physique IV, Colloque C1, 6, 175-185.
Duffrene, L. (1994). "Comportement viscolastique d'un verre silico-sodocalcique dans le domaine
des tempratures intermdiaires." Ph.D. Thesis (in French), Ecole Nationale Suprieure des
Mines de Paris, Paris, France.
Evans, A.G. (1974). "Slow crack growth in brittle materials under dynamic loading conditions." Int.
J. Fracture, 10 (2), 251-259.
Gardon, R. (1980). "Thermal tempering of glass." Glass: Science and Technology, Uhlmann D. R.
and Kreidl N. J., eds., Academic Press, New-York, USA.
Grenet, L. (1899). "Mechanical strength of glass." Bull. Soc. Enc. Nat. Paris, Serie 5, 4, 838-848.
Griffith, A. (1920). "The phenomena of rupture and flow in solids." Philosophical Transactions of
the Royal Society of London, Series A, 221, 163-198.
Guillemet, C., Gy, R., Labrot, M., Sipp, A. and Neuville, P. (1992). "Viscosity, Configurational
entropy and structural relaxation of soda-lime-silica glass." Proc., XVIth International Congress
of Glass, Bol. Soc. Esp. Ceram. Vid. , Madrid, Spain.
Gy, R., Duffrene, L. and Labrot, M. (1994). "New insights into the viscoelasticity of glass." J. Non-
Cryst. Solids, 175, 103-117.
Helfinstine, J.D. (1980). "Adding static and dynamic fatigue effects directly to the Weibull
distribution." J. Am. Ceram. Soc., 63 (1-2), 113-125.
Lawn, B.R. and Marshall, D.B. (1977). "Contact fracture resistance of physically and chemically
tempered glass plates: a theoritical model." Phys. and Chem. of Glasses, 18 (1), 7-18.
18
Michalske, T.A. and Freiman, S.W. (1983). "A molecular mechanism for stress corosion in vitreous
silica." J. Am. Ceramic Society, 66 (4), 284-288.
Narayanaswamy, O.S. (1971). "A model of structural relaxation in glass." J. Am. Ceramic Society,
54 (10), 491-498.
Narayanaswamy, O.S. (1978). "Stress and structural relaxation in tempering glass." J. Am. Ceramic
Society, 61 (3), 146-152.
Narayanaswamy, O.S. and Gardon, R. (1969). "Calculation of residual stresses in glass." J. Am.
Ceramic Society, 52 (10), 554-558.
Redner, A.S. (1995). "Back to basics: nondestructive evaluation using polarized light." Materials
Evaluation, 53 (6), 642.
Rooke, D.P. and Cartwright, D.J. (1976). Compendium of stress intensity factors. Hillington
Press, Uxbridge Middlesex, England.
Schwarzl, F. and Staverman, A.J. (1952)., "Time-temperature dependance of linear viscoelastic
behavior." J. Applied Physics, 23 (8), 838-843.
Sinha, N.K. (1978). "Stress state in tempered glass plate and determination of heat-transfer rate."
Experimental Mechanics, 18 (1), 25-34.
Tool, A.Q. (1946). "Relation between inelastic deformability and thermal expansion of glass in its
annealing range." J. Am. Ceramic Society, 29 (9), 240-253.
Weibull, W.A. (1951). "A statistical distribution function of wide applicability." J. Appl. Mech., 18,
293-297.
19
Wiederhorn, S.M. (1967). "Influence of water vapor on crack propagation in soda-lime silicate
glass." J. Am. Ceramic Society, 50 (8), 407-417.
APPENDIX II. NOTATIONS
The following symbols are used in this paper:
a = crack length
A= Evan's parameter
C
i
= weights in the structural relaxation function
C
p,g
= specific heat of solid glass
C
p,l
= specific heat of liquid glass
G = shear modulus
G
g
= instantaneous shear modulus
H= energy of activation
h
i
= heat transfer coefficients
K = bulk modulus
K
I
= Stress intensity factor (SIF)
K
Ic
= critical SIF
K
Icb
= crack branching SIF
K
e
= deferred bulk modulus
K
g
= instantaneous bulk modulus
K
Ith
= threshold SIF
m = apparent Weibull modulus
m
*
= intrinsic Weibull modulus
M
v
= structural relaxation volume function
20
n = Evan's exponent
P
f
= failure probability
R = perfect gas constant
S = area of the possible location of fracture
S
0
= reference area
T = temperature
t
*
= intrinsic time
T
0
= initial temperature
T
1
-T
2
= temperature step
T
f
= fictive temperature at T
2
t
f
= life time
T
g
= transition temperature
T
ref
= reference temperature
V = instantaneous specific volume
V
, 2
= equilibrium volume at T
2
V
0,2
= volume just after the temperature change
w
ij
= weights in relaxation functions expanded into Prony series
Y = shape factor
= instantaneous thermal expansion coefficient

g
= thermal expansion coefficient of solid glass

l
= thermal expansion coefficient of liquid glass
= thermal conductivity

i
= structural relaxation times
= applied bending stress

0
= apparent Weibull stress
21

0
*
= intrinsic Weibull stress

i
= strength in inert medium

u
= threshold stress (P
f
= 0)

ij
= relaxation times in relaxation functions expanded into Prony series
= reduced time

i
= relaxation functions
22
Figures and tables
List of captions
Fig. 1. Specific volume versus temperature and definition of the fictitious temperature
Fig. 2. Finite element model
Fig. 3. Integral of the difference of principal stresses versus distance from the edge
Fig. 4. Computed stress variation versus time at the edge and in the inner part of the plate
Fig. 5. Computed stress variation versus thickness coordinate at the edge and in the inner part of the
plate
Fig. 6. Four point bending test on small specimens
Fig. 7. Predictions with the Weibull model
Fig. 8. Residual stress parallel to the edge versus edge distance
Fig. 9. Half perspective of the experimental set-up for large specimens
Table 1. Thermoviscoelastic characteristics
Table 2. Viscous and structural relaxations - weights and relaxation times (Tref = 864 K)
Table 3. Thermal conductivity and specific heat
Table 4. Identification of unknown heat transfer coefficients
Table 5. Experimental failure stresses of small specimens
Table 6. Failure stresses of large annealed and tempered specimens (


= 0.05 MPa/s)
23
Specific volume
Temperature
Glass
Liquid
v
1
v
2
T
f
T T
1
T
2
transition range
V
02
V
1
V
2
Fig. 1. Specific volume versus temperature and definition of the fictitious temperature
X
Y
Z
h
1
h
2
h
3
Fig. 2. Finite element model
24
-10
10
30
50
70
90
110
0 10 20 30 40
d = edge distance (mm)
Integral of the difference of principal stresses
Optical measurements
Calculations
T0 = 620C
h1 = 135 W/m
2
.K
h
2
= 115 W/m
2
.K
d
X
Y
Z
A
(MPa)
Fig. 3. Integral of the difference of principal stresses versus distance from the edge
25
-150
-100
-50
0
50
100
0 100 200 300 400 500
Time (s)

zz
(MPa)
Node 1 Node 2
Node 3 Node 4
X
Y
Z
1
2
3
4
Fig. 4. Computed stress variation versus time at the edge and in the inner part of the plate
26
-125
-100
-75
-50
-25
0
25
50
75
-10 -5 0 5 10
x (mm)
zz (MPa)
Edge stress Inner stress
Fig. 5. Computed stress variation versus thickness coordinate at the edge and in the inner part of the
plate
L = 230 mm
19
37.5
polished edge
chamfer
1.5
L/2
Fig. 6. Four point bending test on small specimens
27
0
20
40
60
80
100
20 30 40 50 60
Failure stress (MPa)
Failure probability (%)
tests 0.5 MPa/s
prediction 0.5 MPa/s
tests 0.05 MPa/s
prediction 0.05 MPa/s
Fig. 7. Predictions with the Weibull model
28
-110
-70
-30
10
50
0 50 100 150
Edge distance (mm)
zz (MPa)
-110
-105
-100
0 0.05 0.1
Fig. 8. Computed residual stress parallel to the edge versus edge distance
Fig. 9. Half perspective of the experimental set-up for large specimens
Table 1. Thermoviscoelastic characteristics
29
Young's modulus E Poisson ratio K
e
/ K
g
(2) H / R (5)
g
(9)
l
(9)
70 GPa 0.22 0.18 55000 K 9 10
-6
C
-1
25 10
-6
C
-1
30
Table 2. Viscous and structural relaxations - weights and relaxation times (T
ref
= 864 K)
i w
1i
G
g
(GPa)
1i
(s) w
2i
K
g
(GPa)
2i
(s) C
i

i
(s)
1 1.5845 6.658 10
-5
0.7588 5.009 10
-5
5.523 10
-2
5.965 10
-4
2 2.3539 1.197 10
-3
0.7650 9.945 10
-4
8.205 10
-2
1.077 10
-2
3 3.4857 1.514 10
-2
0.9806 2.022 10
-3
1.215 10
-1
1.362 10
-1
4 6.5582 1.672 10
-1
7.301 1.925 10
-2
2.286 10
-1
1.505
5 8.2049 7.497 10
-1
13.47 1.199 10
-1
2.860 10
-1
6.747
6 6.4980 3.292 10.896 2.033 2.265 10
-1
29.63
31
Table 3. Thermal conductivity and specific heat
Thermal conductivity
(W / m.K, T in C)
Specific heat of liquid glass
C
p,l
(J / kg.K, T in K)
Specific heat of glass
C
p,g
(J / kg.K, T in K)
0.975 + 8.58 10
-4
T 1433 + 6.5 10
-3
T 893 + 0.4 T - 1.8 10
-7
/ T
2
32
Table 4. Identification of unknown heat transfer coefficients
Epibiascope Babinet compensator
Mean measurement (MPa) 120.4 72
Standard deviation (MPa) 4.9 3.3
Apparatus accuracy (MPa) 6 1.2
Identification of h
1
and h
2
(W/m
2
K) h
1
= 135 h
2
= 115
33
Table 5. Experimental failure stresses of small specimens
Failure stress (MPa) first surface finish second surface finish


(MPa/s) 0.05 0.5 0.05 0.5 5
P
f
= 0.2 37.0 41.5 39.9 45.1 45.2
mean 41.2 45.4 50.1 54.7 56.1
P
f
= 0.8 43.7 49.1 57.2 64.0 66.2
34
Table 6. Failure stresses of large annealed and tempered specimens (


= 0.05 MPa/s)
Failure stress (MPa) Annealed glass Tempered glass
Average of tests 31.3 126.8
P
f
= 0.2 30.7 126.7
Prediction mean (P
f
= 0.44) 33 129.0
P
f
= 0.8 35.5 141.5